Rapid urbanization has profoundly impacted social and environmental systems. China is the world’s largest developing country. Projecting future urban expansion in China is critical to alleviate any adverse impacts and achieve sustainable development goals. However, the existing national-scale urban expansion simulations at low or medium resolutions produce significant distortions in urban spatial patterns, limiting the utility of future projections. High-resolution simulation of urban expansion is challenging on a large scale owing to high computational demands. In this study, we used a high-performance cellular automata model (Tensor-FLUS) to simulate the urban expansion of China from 2015 to 2050, at 30 m spatial resolution, under shared socioeconomic pathways. The high-resolution (30 m) urban expansion simulations preserve greater spatial details and avoid 34.07–37.60% underestimation of the urban area, compared with simulation at 1 km resolution. The environmental impact assessments revealed that future urban expansion mainly encroaches on cropland (88.00–88.29%), with a greater likelihood of occupying productive cropland, placing substantial pressure on food production. Although the proportion of occupied natural land is relatively small (11.32–11.60%), newly expanded urban areas will tend to consume woodland and grassland of high ecological value, leading to profound impacts on the ecosystem. In general, the produced high-resolution future simulations can reduce the uncertainty of environmental impact assessment at the national scale. Furthermore, it can provide consistent projection data for research at the provincial or metropolitan scales to support urban planning and local climate change mitigation.

Fine knowledge of the spatiotemporal distribution of the population is fundamental in a wide range of fields, including resource management, disaster response, public health, and urban planning. The United Nations’ Sustainable Development Goals also require the accurate and timely assessment of where people live to formulate, implement, and monitor sustainable development policies. However, due to the lack of appropriate auxiliary datasets and effective methodological frameworks, there are rarely continuous multi-temporal gridded population data over a long historical period to aid in our understanding of the spatiotemporal evolution of the population. In this study, we developed a framework integrating a ResNet-N deep learning architecture, considering neighborhood effects with a vast number of Landsat-5 images from Google Earth Engine for population mapping, to overcome both the data and methodology obstacles associated with rapid multi-temporal population mapping over a long historical period at a large scale. Using this proposed framework in China, we mapped fine-scale multi-temporal gridded population data (1 km × 1 km) of China for the 1985–2010 period with a 5-year interval. The produced multi-temporal population data were validated with available census data and achieved comparable performance. By analyzing the multi-temporal population grids, we revealed the spatiotemporal evolution of population distribution from 1985 to 2010 in China with the characteristic of concentration of the population in big cities and the contraction of small- and medium-sized cities. The framework proposed in this study demonstrates the feasibility of mapping multi-temporal gridded population distribution at a large scale over a long period in a timely and low-cost manner, which is particularly useful in low-income and data-poor areas.

Cellular automata (CA) models have been extensively employed to predict and understand the spatiotemporal dynamics of land use. Driving factors play a significant role in shaping and driving land-use changes. Mining land-use transition rules from driving factors and quantifying the contribution of driving factors to land-use dynamics are fundamental aspects of CA simulation. However, existing CA models have limitations in obtaining accurate transition rules and reliable interpretations simultaneously for multiple land-use simulations. In this study, we constructed a CA model based on a tree-based deep learning algorithm, deep cascade forest (DCF), to improve multiple land-use simulations and driving factors analysis. The DCF algorithm was utilized to mine accurate multiple land-use transition rules without overfitting to improve CA simulation accuracy. Additionally, a novel ensemble mean decrease of impurity (MDI) factor importance analysis method (DCF-MDI), which leverages the cascade structure of the DCF model, was proposed to qualify the contribution of each driving factor to land-use dynamics stably and efficiently. To evaluate the effectiveness of the proposed DCF-CA, we applied the model to simulate land-use distributions and explore the driving mechanisms of land-use dynamics in the Pearl River Delta (PRD), China, from 2000 to 2010. Compared to existing models, the proposed DCF-CA model exhibits the highest accuracy (FoM = 23.79%, PA = 39.77%, UA = 36.35%, OA = 91.50%), which demonstrates its superiority in mining accurate transition rules for capturing multiple land-use dynamics. Factor importance analysis reveals that the proposed DCF-MDI method yields more stable ranking orders and lower standard deviation of contribution weights (<0.10%) compared to the traditional method, indicating its robustness to random disturbances and effectiveness in elucidating the driving mechanisms of land-use dynamics. The DCF-CA model proposed in this study, demonstrating high simulation accuracy and reliable interpretability simultaneously, can provide substantial support for sustainable land use management.

Quantifying woody vegetation cover is essential for understanding vegetation evolution and formulating land use management policies. However, most attention has focused on the retrieval of forest attributes, and it is imperative to enhance the capability to monitor the cover of woody vegetation. This study proposed an adaptive stacking regressor with a self-iterative optimization module to improve the quantification of the fractional woody cover. The adaptive stacking model was trained using synthetic mixing techniques and the 2021 Landsat composite to estimate the proportion of woody cover in Shaanxi Province, China. Results show that: (1) The proposed model exhibits a stable relationship ( R ² =0.92, RMSE =0.1020) between Very High-Resolution (VHR)-derived reference data and the Landsat-based wood cover fraction estimates; (2) The proposed model outperforms other classifiers in heterogeneous region mapping of woody vegetation, followed by Random forest ( R ² =0.82, RMSE =0.1492), Gradient-boosting decision tree ( R ² =0.79, RMSE =0.2550), and Classification and regression tree ( R ² =0.69, RMSE =0.2042) models; (3) Uncertainty analysis based on the Monte Carlo method revealed that the adaptive stacking model improved the overestimations of low cover fractions and underestimations of high cover fractions. The adaptive stacking model is expected to advance the systematic monitoring of biophysical parameters of the land surface.